Considering Uncertain Parameters in Non-Gaussian Estimation for Single-Target and Multitarget Tracking

Consider filtering is an estimation technique that emerged in the 1960s to account for uncertainties in system parameters while simultaneously reducing system dimensionality and (accordingly) the real-time computational cost, along with guarding against issues concerning observability of the parameters. Single-target and multitarget tracking are estimation problems where the dynamics and measurements used for filtering contain uncertain parameters, and the way in which these parameters are handled can drastically impact the performance of the filtering recursion. Traditional consider filters are constructed under the minimum mean square error paradigm rather than the Bayesian framework. The current work develops a Bayesian interpretation of the consider filter, which is then used to derive non-Gaussian single-target and multitarget filtering recursions using Gaussian mixture models. A Monte Carlo analysis validates the statistical consistency of the approach, and a tracking application is presented that d...